The method
How the Comparator works, in five steps
The same five steps for a solar plant, a pump, a heat exchanger or a digester. The Comparator works at one-minute resolution: each one-minute sensor reading is set against the model's value for that minute, and a single sensor with its model is enough to start.
Part of the trial architecture in Namindo's ARENA Launchpad application. webdoli is the planned data chain of the trial: independent logging at each site, the channel register with calibrations, the one-minute Comparator of model against measured, and separate views for the trial team, ARENA's reviewers, the hosts and the public.
In the application, Namindo plans to provide webdoli's services to the trial in kind; the application is under assessment and has not yet been awarded. The Comparator on this page is the same one, offered now to plants and trials of any kind. Trials as a service.
Why the model comes first
A plant's readings move with its duty and its weather. Pressure falls when the speed drops. Solar output follows the clouds. Outlet temperature follows the cooling water. A fixed alarm has to sit outside all of that, so it only trips when the fault is already large.
The model takes the duty and the weather out. It uses the laws the equipment obeys, such as pump and fan curves, the affinity laws, heat and mass balances, the sun's position and the module's temperature response, to say what each sensor should read right now. What is left, the differential, is small and steady while the plant is well.
Why machine learning comes second
Machine learning on raw readings needs years of history and many recorded failures before it can be trusted. Machine learning on the differential needs only a few weeks of normal running, because the physics has already explained most of the movement.
When a flag comes up, you also know where to look. Lower pressure with higher power points to the impeller. Lower output on a clear morning that comes back after rain points to dirt on the panels. A warmer outlet with a higher pressure drop points to fouling. The model says which term has moved.
Where we apply it
In every field Namindo engineers
Namindo designs and reviews energy and process plants. webdoli carries that engineering into the data: the same people who model the plant write the model the data is checked against.
Solar PV and bifacial plants
Every string, against its own weather
- We model
- Front and rear plane irradiance from the measured sky, module temperature, inverter limits and clipping, for fixed racks, single-axis trackers and carports.
- The differential shows
- Soiling and when to clean, shading, string and inverter faults, degradation year on year, and the real rear-side gain. Each lost kilowatt-hour is put under a cause: weather, soiling, clipping, curtailment, outage or fault, and set against the business plan.
Weather and solar resource
Typical years, and how this year compares
- We model
- The clear-sky envelope, neighbouring stations and satellite records, and the site's typical meteorological year.
- The differential shows
- A dirty or drifting pyranometer within a day, and whether a poor month was the weather or the plant.
Pumps, fans and compressors
Wear seen through the duty
- We model
- The maker's performance curve scaled by the affinity laws, efficiency and motor losses, bearing temperature and vibration against speed and load.
- The differential shows
- Impeller wear, recirculation, blocked strainers, bearing and alignment faults.
Heat exchangers, chillers and cooling
Fouling and lost cooling, including data centres
- We model
- Heat balances and exchanger effectiveness, chiller efficiency from load and lift, fan and pump power for the cooling duty.
- The differential shows
- Fouling, refrigerant loss, poor approach temperatures, and cooling energy spent for nothing.
Biogas and digesters
Gas yield against the feed
- We model
- Gas and methane yield from feed rate, solids and retention time, digester heat balance, engine and boiler efficiency.
- The differential shows
- An upset digester before the gas falls away, foaming, gas leaks and engine losses.
Hydrogen, ammonia and process plants
Mass and energy balances that must close
- We model
- Electrolyser stack voltage against current and temperature, mass balances over storage, lines and compressors, and plant energy balances.
- The differential shows
- Stack ageing, small leaks a balance cannot hide, compressor wear and wasted energy.
Live examples
See it work on three plants
Sixty days for each plant, shown hourly so the charts stay readable; the Comparator itself runs on every minute. The data is made up for this page, but the method is the one we run. The model and the envelope learn from the first three weeks. In the fault case a fault starts part way through, and the model is not told.
How the examples work: each plant has a short engineering model (pump curve and affinity laws; irradiance, module temperature and inverter limit; exchanger effectiveness and pressure drop). The differential is averaged over six hours (six daylight hours for the solar plant). The envelope is its spread in each band of duty during the learning window, and a flag needs three hours outside it. On real plants we fit the model to the plant's own test data and use richer learners on the same differential.
How the Comparator differs
Three ways to watch a plant, and why we combine two
Monitoring tools on the market fall into three groups. Each has a strength and a gap. The Comparator joins the physics model to a learned envelope, so it is useful from the first weeks and still explains what it finds.
| Physics model only | Pattern learning on history | The Comparator | |
|---|---|---|---|
| How it works | An engineering model says what the plant should do; anything off is reported | Software learns how the sensors normally move together from months or years of history | An engineering model sets each sensor's expected value every minute; machine learning learns the normal spread of the difference |
| Useful from | Day one | After a long history, ideally with past failures in it | After a few weeks of normal running |
| New duty or weather | Handled, if the model covers it | Often a false alarm, because it was not in the history | Handled by the model; the envelope stays narrow |
| Small drifts | Hidden in model error and sensor offsets | Found, once enough history exists | Found: the learned envelope absorbs the fixed offsets |
| Says why | Yes | Rarely, or through an analyst | Yes: the model shows which term has moved |
| Needs | A good model | A plant historian, many tags and a large licence | One sensor and its model to start; any make of sensor |
No lock-in
Your sensors, your data
Some services only work with their own sensors or their own cloud. The Comparator takes readings from any make, and every series can be downloaded at any time.
Engineer in the loop
Flags read by an engineer
In the full service, every flag is reviewed by a Namindo engineer before it reaches you, with the likely cause and what to check.
Second line kept
Fixed limits stay in place
The Comparator gives the early warning. The maker's limits and the standard vibration and temperature limits stay as the second line, as they should.
We follow the open standards where they apply: IEC 61724-1 for solar plant monitoring (one-minute data, Class A), ISO 20816 for vibration severity, and ISO 17359 and ISO 13374 for the steps of condition monitoring, from data to advice.
Weather models
Typical years from past records, and tracking each new year
Every energy model stands on its weather. We build the weather year for the site, check the site's own weather station every day, and set each new month against the typical one.
The typical year
Built from past records
We take a site's typical meteorological year from public typical-year files, long station records, satellite and reanalysis data, or your own station. It gives hourly sun, temperature, wind and humidity for design and for the model.
Each new year
Tracked month by month
As a plant runs, each month's logged weather is set against the typical year. The plant is judged on the weather it actually had, so a poor month caused by cloud is told apart from a poor month caused by the plant.
The station itself
Checked every day
Each sensor is checked against the clear-sky envelope and its neighbours. A fouled or drifting pyranometer, or a stuck wind vane, is flagged to the field engineer before it spoils a month of data.
We run this for fixed racks, single-axis trackers and sun-following carports, and for new mounting designs before they are built. See the public results for sixteen large solar farm sites on three continents and for Namindo BIPV carports and pergolas in eight Australian capitals.
New service
Reflected typical year, for your site
An hourly typical year of front, rear and ground-reflected irradiance for your location, your mounting and your ground surface, with the energy it gives. Calculated by Namindo, delivered as a file.
What we log
Any sensor, any make
| Quantity | Typical sensors | What we keep |
|---|---|---|
| Vibration | Accelerometers and velocity sensors, wired or wireless, through an edge analyser | Overall velocity and acceleration, peak and crest factor, band energies and bearing defect bands; spectra on request |
| Temperature | RTDs, thermocouples, infrared and winding sensors | One-minute mean, minimum and maximum |
| Pressure and flow | Transmitters on 4 to 20 mA or digital; magnetic, ultrasonic and Coriolis meters | One-minute statistics and totals |
| Torque, speed and power | Torque transducers, encoders, drives and meters | Shaft and electrical power, efficiency, load |
| Electrical | Power meters, inverters, drives, current clamps | Power, energy, current, voltage, power factor; harmonics where available |
| Irradiance and weather | Pyranometers, reference cells, albedometers, weather stations | One-minute statistics with daily quality checks |
| Process | Level, gas composition, pH, humidity | One-minute statistics, alongside the equipment data |
We are not tied to a make. Readings can come from a data logger, a PLC, a drive, a SCADA or historian export, an IoT gateway or a CSV file. Where a site speaks Modbus, OPC UA or MQTT, a small gateway at site turns it into HTTPS for us. Nothing on your control network has to accept a connection from outside.
The service
Use the Comparator yourself, or let us run the watch
Companies anywhere can use it. Your sites can be in one country and your engineers in another. We work from Australia, Indonesia and India.
Platform
You send, we keep
Logging from your own loggers, a register of every channel with its calibration, charts, CSV and OpenOffice downloads, and named logins for your team, your client and your auditor.
Platform and models
We model your plant
Namindo engineers write the model of each machine or plant from its data sheets, test curves and drawings, learn the envelope from your normal running, and flag differentials by email.
Full service
We run the watch
We specify and fit sensors and cellular loggers, keep the calibrations, review every flag and send a monthly health report with the data behind it.
Your data stays yours
- Download any series at any time as CSV or OpenOffice, at one-minute, hourly or daily steps.
- Records cannot be edited. A wrong reading is voided with a reason and stays on record.
- Each logger has its own key, and we can switch it off at once.
- Logins by email, with read-only accounts for clients and reviewers.
In-house examples and track record
- In-house example: the comparators planned for the field trial in Namindo's ARENA application, where measured solar output will be set against the model at four sites, including two demonstration units.
- In-house example: Namindo's carport and pergola modelling and sixteen-site solar farm modelling, hour by hour on public typical years.
- Remote condition monitoring of plant assets: diagnostics to prognostics on a remote island plant (talk, 2021).
- Machine to database: DCS, PLC and warehouse data in one record for a large ammonia complex (talk, 2018).
Talk to us about your plant
Tell us what you run, what it measures today and what you would like to know sooner. Email info@webdoli.com or call +61 426 087 878.